FYI.

This story is over 5 years old.

Tech

This Twitter Bot Is Generating Jokes and It's Funnier Than You

Headlinertron is processing comedy specials to create its own jokes.
Ashwin Rodrigues
Brooklyn, US

Open-mic comedy is perhaps the most fickle form of art, both as a performer and a spectator. When it’s bad, it’s physically jarring. When it’s good, it’s inspiring.

The Headlinertron, the creation of Florida-based comedian CJ Hernandez, is a Frankenstein-like Twitter bot that captures both the awkward incoherence and flashes of brilliance associated with open-mic standup comedy.

As a fan of stand up comedy, and a comedian himself, Hernandez has collected a large number of comedy special transcripts, from comedians like Dave Chappelle, John Mulaney, Sarah Silverman, and Chris Rock.

Advertisement

“I've always found it really interesting to look at stand up comedy on paper, it gives a certain insight into how the pacing works sometimes,” Hernandez told me over email.

His collection of standup transcripts exceeds 300,000 words. Feeding chunks of transcripts into a predictive text generator called Botnik, Hernandez creates sometimes nonsensical, sometimes hilarious bits which he posts on Twitter.

“When a comedian starts out, they start out by emulating their favorite comic. That's literally what this bot is doing. It's very honest to the process of many new comedians,” Hernandez said.

Strictly speaking, the Headlinertron is not a “bot.” It’s more a role reversal between human and machine. Botnik, with comedy special transcripts as its source of knowledge, generates the creative output. Hernandez simply ensures the sentences make sense, and adds punctuation, before manually loading the jokes into the social media managing platform Buffer, where he schedules tweets for the Headlinertron Twitter account.

To keep the process more machine than man, Hernandez prefers to pick the first suggestion that continues the sentence in a grammatically correct way.

After prompting the Headlinertron with “Steve Jobs,” I found the machine holds some troubling views.

“If it was purely robotic, it wouldn't be funny. But if it was too human, it wouldn't be interesting. No one wants to see a random mishmash word salad that a pure markov chain generator's putting out, but no one wants to see some dude pretending to be a robot either,” Hernandez said.

Advertisement

An example of a longer Botnik-generated text string. Image: CJ Hernandez

By correcting grammatical errors or inserting punctuation, I wondered the ethical implications of running a joke generator. Was this “cheating?” What does cheating even mean, in this context?

To understand the unspoken rules around predictive-text generation, I spoke to co-founder of Botnik and former ClickHole writer, Jamie Brew. While he’d seen similar projects, like Botnik’s own Seinfeld predictive keyboard, Brew said he’d never seen one at the scale of the Headlinertron.

I asked Brew, in terms of the human interference, how much interference is “allowed?”

“We talk about this a lot in the internal botnik channels,” Brew said.

“Personally, when I'm using it, I tend to only try for sequences of words that come directly from the interface,” Brew said, but mentioned others are more liberal, allowing for the re-shuffling of suggested words, while others take an extreme approach to use Botnik’s first suggestion without any further editing.

Before our call ended, Brew assured me, “CJ [Hernandez] is innocent.”

Though Hernandez says his formal computer science experience is “practically none,” he’s considering a more sophisticated future for the Headlinertron.

“I would like to do some research and try doing a real neural network type of approach to see if it would work, but I don't know much about it so I'm kind of learning from scratch,” he said. Hernandez has plans to bring his creation to a real stage. “I also did order a robot mask on Amazon, and I'm gonna do (the robot jokes) live. That should be fun. “